Questions about the output

[TinaShi]

Thanks for your help with the code. For the annotation file in the samples folder, after running 100,000 iterations with max model size == 100, I get an output as below.

Here I have two more questions about the output. The first model contains main effects -1;1;33;73;523. What does the "-1" mean here? Second, why does all the log-likelihood equal to 0 for this case? It looks a little strange to me. Do you still remember the possible reasons that lead to this result?

[gchen98]

Hi there,

-1 just represents the intercept, 0 would be the first variable in your model. Not sure about the log likelihoods of zero..

If you see the samples/endophenotypes folder and run the shell script you should see models.out look something like:

MODEL MAINEFFS SIZE LOG-LIKELIHOOD 1 -1;0;1;2;3;4 6 -11079.4 2 -1;0;1;2;3;4 6 -11066.8 3 -1;0;1;2;3;4 6 -11074.9 4 -1;0;1;2;3;4 6 -11088.6 5 -1;0;2;3;4;10 6 -11077.3 6 -1;0;2;4;9;10 6 -11065.3 7 -1;0;2;4;9;10 6 -11065.2 8 -1;0;2;4;9;10 6 -11063.4 9 -1;2;4;9;10;11 6 -11065.6 10 -1;2;4;9;10;11 6 -11065.5 11 -1;2;9;10;11;15 6 -11061 12 -1;2;9;10;11;15 6 -11060.9 13 -1;2;9;10;11;15 6 -11065.1 14 -1;2;9;10;11;15 6 -11060 15 -1;2;9;10;11;15 6 -11060 16 -1;2;9;10;15 5 -11061.8 17 -1;2;9;10;15 5 -11059.8 18 -1;2;9;10;15 5 -11061.8 19 -1;2;9;10;15 5 -11058.8 20 -1;2;9;10;15 5 -11061.8 21 -1;2;9;10;15 5 -11057.8

On 05/30/2017 02:14 PM, TinaShi wrote:

Thanks for your help with the code. For the annotation file in the samples folder, after running 100,000 iterations with max model size == 100, I get an output as below. screen shot 2017-05-30 at 2 03 07 pm https://cloud.githubusercontent.com/assets/12648659/26605202/f45803ae-4541-11e7-82b5-3eda137162cb.png Here I have two more questions about the output. The first model contains main effects -1;1;33;73;523. What does the "-1" mean here? Second, why does all the log-likelihood equal to 0 for this case? It looks a little strange to me. Do you still remember the possible reasons that lead to this result?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/gchen98/pimsa/issues/2, or mute the thread https://github.com/notifications/unsubscribe-auth/AKfCuu130o3iRdCnfPwVjz8HHhrEDSrCks5r_IahgaJpZM4Nq2-M.

[TinaShi]

Hi, Thanks for the information. I notice that in the endophynote dataset you used logistic regression, while in the annotation dataset the logistic regression is suppressed, since the phenotype dataset contains a continuous variable, I suppose you have used a linear regression here. In the step(2) of the appendix of your paper, you specified the log likelihood function (A1) for logistic regression, I suppose we need to change the formula when we use linear regression instead. Since we need this value in step (8) formula (A9). I think you probably have provided the formula for linear regression as well, we just need to show it when logistic is set to be false in the pima.xml. I do not think the 0 values are used later on, otherwise (A9) is undefined when the denominator is 0, and I was not able to get the chain going.

These are all my guesses, please let me know if you can provide more insight on this topic. I will also learn more about your code to understand it better.

[TinaShi]

Hello, my last guess is not right. I just noticed a strange pattern, that is the output files for beta and prior variances are all 0! It starts to show this pattern from the first model. I think it might have something to do with logistic regression, but I am not sure.

[TinaShi]

I just tried to simulate binary variables of the same size for annotation dataset and use logistic regression, unfortunately the same problem still exists. I compare the pimsa.xml in annotation and endophenotype, the difference is that use_endoprior is set to be true in endophenotype and marginal_prior is set to be true in the annotation dataset. I suspect the problem exists in the marginal_prior related files. After searching the file, I notice code like this "if (settings->marginal_prior){logLike = 0;", " if (settings->marginal_prior){ 662 //getRandomBetas(ms,currentModel->betas); 663 logL = 0.;" in pathwaysearch.cpp file. But I do not understand why this is the case. Can you help me to figure it out? I have been stuck on this project for quite some time, my application has the same data properties as your annotation dataset. It will be of great help if you can help fix the problem. Thanks in advance.